1. Reminder: Please use our affiliate links for holiday shopping!

What if there was a 20mp BSI Micro 4/3 sensor?

Discussion in 'Olympus Cameras' started by Luckypenguin, Mar 30, 2016.

  1. For a while I've been meaning to do a comparison between a Micro 4/3 camera and one with a state-of-the-art APS-C sensor, using lenses of the same focal length and cropping the APS-C image down to equalise the framing. To do this test I have used an Olympus E-M1 and a Samsung NX500. When cropped down to match the framing and aspect ratio of a Micro 4/3 camera, the resolution of the NX500 image is approximately 20mp (down from 28mp). The purpose of this was to see whether there would be a tangible benefit by creating a Micro 4/3 sensor using the latest in technology.

    This is how the two sensors compare by numbers alone:

    [​IMG]
    Source: Samsung NX500 vs Olympus OM-D E-M1 | DxOMark

    Now keep in mind that by cropping the Samsung sensor it will automatically have a smaller dynamic range and it's low light noise threshold will be reduced, since the noise generated per pixel doesn't change but there are less pixels overall for the noise to hide in. What that means is that a Micro 4/3 sized version of the Samsung sensor wouldn't score as highly as the full sized version does.

    My test was to take an image of a high dynamic range scene with each camera at their base ISO; 100 for the NX500 and 200 for the E-M1. While the E-M1 does have a simulated ISO 100 (LOW) setting, this is just an overexposed ISO 200 image that has been pulled back by one stop that gives less shadow noise but also less highlight dynamic range. If I was to use the LOW setting on the E-M1 it could be argued that I could do the same thing manually with the Samsung to create a simulated ISO 50 setting.

    The lenses used in this test were the 20mm f1.7 pancake on the E-M1 and the 20mm f2.8 pancake on the NX500.

    E-M1 unedited
    [​IMG]

    E-M1 edited
    [​IMG]

    NX500 edited and cropped to match framing.
    [​IMG]


    Comparisons of 100% crops at various locations

    [​IMG]

    [​IMG]

    [​IMG]

    [​IMG]

    [​IMG]

    [​IMG]

    [​IMG]

    [​IMG]

    [​IMG]

    For me the difference comes down to the E-M1 having more shadow noise, and of course the slightly higher resolution of the Samsung even when cropped. In this case I feel that the key advantage enjoyed by the Samsung comes from it having a lower base ISO.


    My second test was performed in low light at ISO 6400 using each manufacturer's respective 45mm f1.8 lens.

    E-M1
    [​IMG]

    NX500
    [​IMG]


    Comparison of 100% crops

    [​IMG]

    [​IMG]

    [​IMG]

    [​IMG]

    Again I think that this a small win to the cropped NX500 sensor, most obviously with less colour bleeding into the blacks.

    I apologise if this wasn't the most exhaustive comparison but it has been enough to show me that there would be a small benefit to applying the latest in sensor technology to Micro 4/3 cameras. Maybe Olympus could buy up the remaining stock of NX1/NX500 sensors and file them down to size :).
     
    Last edited: Mar 30, 2016
    • Informative Informative x 6
    • Like Like x 5
  2. Nawty

    Nawty Mu-43 Regular

    83
    May 1, 2015
    Thanks for posting, for me this just illustrates the point that if m43 shooters are expecting some magic night and day differences from a newer BSI sensor then they will be sorely disappointed.
     
    • Agree Agree x 2
  3. wjiang

    wjiang Mu-43 Hall of Famer

    Not night and day but then neither is FF if we're talking RAW at only ISO 6400. The difference in colour fidelity is pretty huge though, and for astrophotography this could put m4/3 over into very usable territory. As it is I'm scraping by with ISO1600/3200 and stacking to control the noise.
     
  4. Klorenzo

    Klorenzo Mu-43 All-Pro

    Mar 10, 2014
    Lorenzo
    Thanks for the comparison. To be honest in the non flat elements I prefer the grainy/noisy rendering of the E-M1, it looks more real (even if the real object may be different).

    Did you use default denoise for both cameras? PS is known to have bad default settings for the OM-D.

    I really do not want to hijack the thread but cropping, by itself, does not affect any image parameter. If I cut an image in half of course nothing changes in each of the two halves. The bigger sensor allows me to use more surface to record the subject but if the extra surface is mostly on the sides my subject will not get more "recording surface". I'll just get extra elements on the sides of the subject and these do not improve the recording of the subject.
    Take this extreme example: take four m43 sensor and stick them side by side in one single long line, like train wagons. The DxO stats probably rise to full-frame levels (I suppose) but when you take an headshot in landscape orientation nothing has changed.

    So depending on how you framed the shot, how much squared millimeters of the sensor you used for the window for example, the crop advantage may be more or less relevant.
     
  5. pdk42

    pdk42 One of the "Eh?" team

    Jan 11, 2013
    Leamington Spa, UK
    That's an interesting test. Thanks for doing it. For me, this is what I conclude:

    - The Samsung shows more detail in the low ISO shot (e.g. the grain on the wood)
    - The Samsung does notably better in colour handling, contrast and bleed on the high ISO shot
    - The differences aren't enormous but they are readily visible on 100% crops
    - The differences on a well-exposed and "simple" shot are probably invisible in most practical ways (e.g. a print or a non-4k screen)
    - The differences probably would matter if a lot of PP were needed or if there was a lot of cropping to be done
    - A 20Mp BSI u43 sensor would be well worth having!

    Please Olympus, could you arrange this on the E-M1ii?
     
  6. Moula

    Moula Mu-43 Regular

    60
    Mar 9, 2016
    Maybe 12 Mp?
     
  7. ijm5012

    ijm5012 Mu-43 Hall of Famer

    Oct 2, 2013
    Pittsburgh, PA
    Ian
    Nicely done @Luckypenguin@Luckypenguin, thank you for your work!

    I think this shows that there still is room for improvement with a m43 sensor, mainly at the low-end of the sensitivity range where if they could implement a native base ISO 100, it would yield better noise characteristics and dynamic range compared to the current native base ISO of 200. That would be huge for me when shooting landscapes when I have to push the shadows in post, but am limited by the amount of noise that appears.

    I think there are benefits at 6400 as well, not only with the color bleeding but also the noise characteristics. There's a bit more detail in the NX500 shot vs the E-M1 shot.

    Honestly, I would love for a new 16MP sensor that performed like the above NX500 sensor. Shift the sensitivity down a stop, so instead of 200-25600, make it 100-12800 (native), and it may actually be the first camera body I wouldn't wait to buy used, haha.
     
    • Agree Agree x 2
  8. pdk42

    pdk42 One of the "Eh?" team

    Jan 11, 2013
    Leamington Spa, UK
    It's interesting to speculate what Olympus will do with the E-M1ii as regards sensor sourcing. The new 20Mp Sony sensor used in the PenF doesn't have PDAF. I'm not knowledgeable enough to know whether this could be added via a custom filter array and image processing software - but if not, then it would indicate a new sensor is needed. AFAIK, the Panasonic GX8 doesn't use PDAF so I'm wondering if Panasonic would be a potential source or not.
     
    Last edited: Mar 30, 2016
  9. ijm5012

    ijm5012 Mu-43 Hall of Famer

    Oct 2, 2013
    Pittsburgh, PA
    Ian
    That's an interesting point. Panasonic was the source for the E-M1's sensor, and you're correct that Sony has never made a m43 sensor with PDAF (maybe to not compete against their APS-C E-Mount cameras? Who knows why...). The GX8 doesn't use PDAF either, as it uses Panasonic's DFD system which is CDAF reliant.

    The sensor in the E-M1 II should be all new, it's just a matter of who will supply it.
     
  10. Growltiger

    Growltiger Mu-43 Top Veteran

    644
    Mar 26, 2014
    UK
    Perhaps Olympus will remove PDAF support from the EM-1 ii, as a trade-off, in order to get a better sensor. They could say they have done what was needed to help people with the old lenses with the E-M1, but now it is time to move on.
     
  11. Turbofrog

    Turbofrog Mu-43 Hall of Famer

    Mar 21, 2014
    Very interesting. I think the E-M1 shots look better in base ISO due to my preference for sharp detail over smoothness. The Samsung versions actually look like they could be noise-reduced versions of the Olympus files.

    But I do think that the Samsung is noticeably superior at ISO 6400, albeit not massively so. But distinctly more detail AND less noise.

    I'm a bit confused when you say that cropping the sensor would reduce in less dynamic range. I may be misunderstanding the physics of dynamic range, but I thought it had to do with the saturation limit / full-well capacity of the individual photosites. So the actual size of the sensor is kind of irrelevant to dynamic range, it just has knock-on effects in determining the specific architecture of the sensor (i.e. the photosite density, ADC design, etc...) which DO affect dynamic range. This seems to be borne out by the Sony 1" sensor having a similar dynamic range to the 16MP M4/3 sensors, which in turn have a similar dynamic range to Canon's full-frame sensors. In each case there is a strong correlation in ISO performance between the physical size of the sensors, but the DR is completely separate. Am I mistaken?
     
    • Agree Agree x 1
  12. While it is true that cropping an image has no impact on the remaining image at a pixel level, a greater level of magnification will be required to achieve the same display output size as the original and therefore the threshold at which noise becomes objectionable becomes lower. Noise increases both when the overall signal amplification (ISO value) is increased and also when shadows are pushed. If the lower limit of sensor dynamic range is defined when texture can no longer be discerned, this lower limit should be different for a smaller (cropped) image that needs to be magnified further.

    The other variable here is the relative sharpness of the two lenses. The 20mm f2.8, while still a decently sharp lens, is the softest of my three Samsung lenses (the other two being the 30mm f2 pancake and the 45mm f1.8). It is also designed for an APS-C sized sensor so cropping it's images to simulate a Micro 4/3 sensor is asking it to resolve higher than the sensor that it was designed for.

    None of the images in the above comparison have any noise reduction applied to them by me in Lightroom, but I don't know whether either camera has any "baked in" to the raw files.
     
    Last edited: Mar 30, 2016
  13. bikerhiker

    bikerhiker Mu-43 All-Pro

    Dec 24, 2013
    Canada
    David
    Would it be better to upsample the 16MP E-M1 files to native 28MP with the correct aspect ratio using Genuine Fractals and present an accurate representation?
     
  14. Perhaps if I was trying to directly compare the E-M1 to the NX500 in all it's full APS-C sized glory :). In that case I would also need to use different focal lengths for each camera to achieve the same framing.

    The aim here was to create a theoretical Micro 4/3 sensor by cropping the images from a camera with a larger sensor that has all the latest bells and whistles with respect to sensor technology, and then compare those to the same images taken with an existing Micro 4/3 sensor.
     
    • Agree Agree x 2
  15. b_rubenstein

    b_rubenstein Mu-43 All-Pro

    Mar 20, 2012
    Melbourne, FL
    Making a comparison to a Samsung sensor isn't very useful, because if there was a 20mp, BSI 4/3 sensor it would be from Sony and exhibit difference performance. This is just a pub thread.
     
  16. Turbofrog

    Turbofrog Mu-43 Hall of Famer

    Mar 21, 2014
    Fair enough. I totally agree that resolution and sensor size come into the play in terms of minimizing the impact of noise. It's part of why A7r (original model, not Mk. II) has significantly better ISO performance than the A7 and A7 II despite having 50% more pixels, it means you can downsample the image much more cleanly. But my general experience with limited dynamic range is in clipped highlights, not muddy shadows. Obviously it's two sides of the same coin, and the solution to that problem is to significantly (i.e. ~3 stops) underexpose your image and then pull up the shadows to retain tonality in highlights, which reveals noise in the shadows. But I guess this discussion is also why it's difficult to come up with any universal definition of what dynamic range really is besides an engineering definition associated with the full-well capacity as I mentioned in my first post (this is what DXO tests, I believe). Other sites, like photonstophotos.net (Bill Claff's site, mentioned previously in this thread) do it a bit differently, since they also take into account SNR-ratio and circle-of-confusion, which somewhat normalizes for sensor size and the magnification thereof, as you are referring to.
     
  17. Turbofrog

    Turbofrog Mu-43 Hall of Famer

    Mar 21, 2014
    I don't think it's "just a pub thread" at all. The NX500 sensor is up there with the best APS-C sensors on the market right now (give or take), so it's a very useful evaluation of how the different technologies stack up vis-a-vis sensor size and output resolution.
     
    • Agree Agree x 1
  18. bikerhiker

    bikerhiker Mu-43 All-Pro

    Dec 24, 2013
    Canada
    David
    Imaging Resources actually did a test between the Samsung NX500 against an E-PL7 @ ISO 1600 and concluded a little differently. That is, despite the BSI sensor, the E-PL7 still performs quite well and in fact slightly BEAT the NX500 in detail retention.. Take a look below of 2 images I downloaded from Imaging Resources in RAW format and then processed with DXO Optics Pro Prime. The E-M1 was then up-sampled using Perfect Resize to 6480x4320. Genuine Fractals is best to use for image quality comparison I find.. You might be shocked to what you see..
    Left image: E-M1 up-sampled via Perfect Resize
    Right image: Samsung NX-500
    Basic RAW conversion ONLY. No sharpening or other things added yet!

    Screen Shot 2016-03-30 at 8.47.29 AM. Screen Shot 2016-03-30 at 8.49.38 AM.

    These images are shoot @ ISO 6400. Post processing, the E-M1 has about 1 stop more shot noise against the NX-500. Obviously, because the size of the sensor is smaller than APS-C, it is more noisier. However, when I spoke with the top end people at Olympus, they told me that sensors alone isn't going to get you the ultimate best image quality the same as a Ferrari isn't going to go fast without a capable race driver like Michael Andretti. That race car driver is the "Imaging Processor Algorithm" and Olympus has many many decades of that. It is the balancing of color, texture, detail and noise that ultimately make the best image!
    In Olympus terms, there are 3 components that make good image quality. Good sensor + Imaging Processor Algorithm + good Glass (Zuiko) = good image quality. Too many people still emphasize on noise, noise and NOISE but noise itself is just one component of the equation!

    As you can see here, it's clear that the E-M1 files are very malleable. And I've noticed too that top end models from Olympus seemed to have better imaging processing algorithm than the lower end models. This is true between my Olympus E-5, which is again very malleable in getting the best image quality as opposed to my former E-PL1 which has the same sensor as the E-5. What Robin Wong said a while back was absolutely right! That is, the E-5 files are clearly better than the E-PL1 despite what DPReview and DXO Mark tells otherwise.

    To me, the E-M1 holds better detail and color retention than the NX-500. It's after all Olympus PRO model. It might still show some noise, but the noise pattern is very acceptable when I print.

    The reason I chose ISO 6400 is because the E-M1 still has "enough" photographic dynamic range to produce a good quality image. Anything higher than ISO 6400, the APS-C and full frame will win because it has 1 or 2 stops more dynamic range and also in the case of Nikon possess the same decades knowledge of imaging processing know how as Olympus. This is why it's important to understand your shutter speed operating conditions and why if you want to step up from the E-M1, the Sony A7R or A7RII or Nikon D810 or D750 would give you a more noticeable return in both low ISO performance and higher ISO performance.
     
    Last edited: Mar 30, 2016
  19. bikerhiker

    bikerhiker Mu-43 All-Pro

    Dec 24, 2013
    Canada
    David
    Currently, the Nikon D7200 has the "best" APS-C sensor on the market today. It's approaching close to about 1/2 stop to a D750, but your example is valid as it clearly showed and I had said before that BSI alone isn't the answer to get cleaner files and isn't the elixir people think it is..
     
  20. Shouldn't the objects on each side display at the same size if the E-M1 has been upressed to the same horizontal resolution as the NX500?